larsgregori.de
I went to bed. 53 minutes later, I had a working NFL game prediction microservice. The Ralph Wiggum paradigm turned a concept document into a deployed application with ML predictions.
Theory said AI-generated tests are circular. Now let's get empirical. I ran mutation testing on vibe-coded tests and found an 85% mutation score. That means 15% of bugs would slip through. Spec-first tests hit 92%. Property tests found 11 real bugs that all example tests missed entirely.
Using a different LLM for testing sounds like it would create independence. It doesn't. The moment any model sees your implementation, it reasons about what the code does, not what it should do. Real independence requires different information sources, not different systems.
Vibe coding's dirty secret - AI-generated tests only prove the code does what the AI thought it should do. At 39C3, security researcher Johann Rehberger showed it's worse than that. Prompt injection attacks can turn your AI coding assistant into a ZombAI, and the tests will still be green.
The AI boom feels like déjà vu for anyone who's weathered past tech bubbles. While media headlines promise sentient machines and mass disruption, real-world adoption tells a slower, more grounded story. This article compares the current AI hype cycle to previous technology frenzies, critiques inf...
Ever wondered why AI gives you meeting dates from 2023 when you ask for "next week"? Dive into Model Context Protocol (MCP) — Anthropic's game-changing approach that connects LLMs to real-world data. I tested three different AI models with the same scheduling request and built several MCP service...
AI meets FPGA! Used an LLM to generate an XOR neural network on a Cyclone 10 FPGA — Verilog, simulation, training, and Arduino integration. A fun journey into AI-powered hardware design!
Explore how I used AI to generate honeycomb SVG patterns in Python, overcoming challenges and refining designs. Plus, my take on whether AI will replace our jobs—it’s more of a partnership than you think.
Exploring the intersection of AI, GPIO, and keyboard design, I built a five-key keyboard prototype powered by neural networks—an oversized but rewarding challenge that combined engineering and creativity.
During our recent xLSTM Meetup, Professor Sepp Hochreiter presented the latest advancements in the xLSTM architecture, showcasing its potential to revolutionize the field of natural language processing.
How I used the GraphRAG paper to enhance structured product data for televisions.
Starting with a small web3 nodejs example and ending up with an ERC20 LARS token.
I want for my iOS apps an own icon which I create with app-icon.
One of my Corona project … my first LinkedIn Learning course is online
.gitignore entries depend on the programming language, development environment and operating system and gitignore.io can help.
In this blog post I describe how I've created a Dockerfile for gnu-poke.
In this blog post I describe how to extend the size of Docker on an EC2 instance.
In this example I split a webservice into two applications and merge them together with Nginx. I also use docker-compose.
docker-compose defines and executes a microservice application that is structured as a multi-container docker application.
If the Raspberry Pi should say goodbye, I only need to know how to start the Docker image.